After looking at how work was actually happening… and what broke when AI tried to scale inside it…
The question changed.
It was no longer:
How do we use AI better?
It became:
What would a workflow need to look like for AI to actually work inside it?
What Needed to Change
The issue wasn’t a single tool, workflow, or use case.
It was how everything connected.
Across both projects, the same gaps showed up:
Work that looked repeatable, but wasn’t structured
AI that could assist, but not operate independently
Decisions that lived in people, not systems
Context that had to be rebuilt every time
The system wasn’t broken.
It just wasn’t designed to carry thinking forward.
The Shift
Instead of trying to improve outputs, the focus moved to designing the layer underneath them.
Not:
better prompts
more automation
faster execution
But:
clearer structure
visible decision points
defined loops instead of one-off tasks
systems that hold context over time
This is where the work started to change.
The Model
Instead of thinking in campaigns, tasks, or outputs, the workflow was reframed as a loop:
Input → what is coming in
Structure → how it is organized
Decision → what needs human judgment
Support → where AI can assist
Output → what gets produced
Feedback → what gets learned
Then back again.
Not a straight line.
A system that carries itself forward.
Where AI Actually Fits
AI didn’t become the system.
It became one part of it.
Specifically:
processing large amounts of information
surfacing patterns that would be hard to see manually
supporting repeatable decisions
But it still depended on:
clear inputs
structured context
defined boundaries
Without those, it drifted.
With them, it became useful.
What This Changed
The goal was no longer to:
use AI inside existing workflows
It became:
design workflows that AI can actually operate inside
That shift changes everything.
Because now:
work becomes easier to repeat
decisions become easier to track
systems become easier to improve over time
And AI stops feeling unpredictable.
What This Shows
Most teams are trying to layer AI on top of work that was never designed for it.
That’s why it feels inconsistent.
That’s why it doesn’t scale.
That’s why it requires constant intervention.
This isn’t a tool problem.
It’s a structure problem.
Pan:
This is where the system becomes visible.
In the first case, the work depended on human coordination.
In the second, the AI exposed the limits of that structure.
Here, the system is being redesigned to hold both.
Not human or AI.
Human + AI, operating within a shared structure.
That is the difference between using tools and designing systems.
Try this with your own workflow
Take something you do every week.
Map it out as:
Input
Decision
Output
Then ask:
Where does this break if I try to repeat it?
That’s usually where the system needs to be built.